The developmental trajectory of fronto‐temporoparietal connectivity as a proxy of the default mode network: a longitudinal fNIRS investigation

Abstract The default mode network (DMN) is a network of brain regions that is activated while we are not engaged in any particular task. While there is a large volume of research documenting functional connectivity within the DMN in adults, knowledge of the development of this network is still limited. There is some evidence for a gradual increase in the functional connections within the DMN during the first 2 years of life, in contrast to other functional resting‐state networks that support primary sensorimotor functions, which are online from very early in life. Previous studies that investigated the development of the DMN acquired data from sleeping infants using fMRI. However, sleep stages are known to affect functional connectivity. In the current longitudinal study, fNIRS was used to measure spontaneous fluctuations in connectivity within fronto‐temporoparietal areas—as a proxy for the DMN—in awake participants every 6 months from 11 months till 36 months. This study validates a method for recording resting‐state data from awake infants, and presents a data analysis pipeline for the investigation of functional connections with infant fNIRS data, which will be beneficial for researchers in this field. A gradual development of fronto‐temporoparietal connectivity was found, supporting the idea that the DMN develops over the first years of life. Functional connectivity reached its maximum peak at about 24 months, which is consistent with previous findings showing that, by 2 years of age, DMN connectivity is similar to that observed in adults.

the most well-known and most studied resting state networks (Raichle, 2015;Sporns, 2010). The DMN is composed of the medial prefrontal cortex (mPFC), the precuneus, the posterior and anterior cingulate cortex, the inferior parietal lobe (IPL), the medial temporal lobe and the temporoparietal junction (TPJ; Davey, Pujol, & Harrison, 2016;Greicius, Krasnow, Reiss, & Menon, 2003;Harrison et al., 2008;Mars et al., 2012;Molnar-Szakacs & Uddin, 2013;Raichle, 2015;Schilbach, Eickhoff, Rotarska-Jagiela, Fink, & Vogeley, 2008;Sporns, 2010). The importance of the DMN is underlined by several recent studies that have found that changes in the connectivity strength in this network are related to many psychopathologies (Broyd et al., 2009) and Alzheimer disease (Greicius, Srivastava, Reiss, & Menon, 2004). Furthermore, adult studies on the DMN suggest that this network is an "intrinsic system" that deals with self-related signals and self-processing (Golland, Golland, Bentin, & Malach, 2008). In fact, areas that are activated during self-processing tasks show extensive overlap with the regions belonging to the DMN (Buckner & Carroll, 2007), and neuroimaging studies have shown that the DMN activity is positively correlated with participant reports of mind wandering and self-related thoughts (Mason et al., 2007;McKiernan, D'Angelo, Kaufman, & Binder, 2006). Given the crucial role of the DMN is thought to play in self-processing, it has been suggested that the gradual development of this functional network also supports the emergence of self-awareness in the first years of life (Gao, Lin, Grewen, & Gilmore, 2016). Consistent with this view, it has been shown that the mPFC, a core region of the DMN, is more activated in response to self-focused stimuli rather than externally-focused stimuli (Xu et al., 2017) and to hearing one's own name rather than another's names (Imafuku, Hakuno, Uchida-Ota, Yamamoto, & Minagawa, 2014) even before the first year of life.
The first study that explored resting-state networks in the infant brain used fMRI with sleeping infants between 4 and 9 months (Fransson et al., 2007). This work showed evidence for the presence of visual and primary sensorimotor networks from birth, results that have since then been replicated several times (Gao, Alcauter, Smith, Gilmore, & Lin, 2015;Lin et al., 2008;Liu, Flax, Guise, Sukul, & Benasich, 2008). However, Fransson et al. did not find evidence for temporal synchronisation in core regions of the DMN before the first year of life (Fransson, Åden, Blennow, & Lagercrantz, 2011;Fransson et al., 2007Fransson et al., , 2009). The early maturation of the primary sensory networks is thought to indicate that primary sensory functions, such as vision and touch, are in place from very early in life (even though functional networks also undergo significant change over the course of development, and adapt to the acquisition of new skills [for example see Marrus et al., 2018]). In comparison, the slow integration of regions belonging to the DMN in a unique network during the first years of life might be consistent with the gradual emergence of more advanced social cognitive abilities (Gao et al., 2016). More recent studies have discovered precursors of the DMN even before the first year of life. For example, functional correlations between core regions of the DMN were found in 4-month-old infants (for instance between the posterior cingulate cortex and the TPJ), but at this age, there was no significant correlation between the time series of the posterior and the anterior components .
While short-separation connectivity decreases with age, functional connectivity between more distant areas tends to increases with age, consistent with the idea of a gradual long-range integration within the DMN . Gao et al. showed precursors of a primitive DMN even at 2 weeks of life, and they demonstrated that by 2 years of age the DMN is functionally similar to that observed in adults (Gao et al., 2009).
All the infant studies mentioned above acquired resting-state data with fMRI in sleeping participants, while resting-state studies on adults usually acquire data on awake participants, who are typically asked not to think about anything in particular. However, connectivity measured during sleep does not display the same patterns of connectivity measured during wakefulness (Horovitz et al., 2009).
Additionally, sleep stages have an effect on estimates of functional connectivity (Mitra et al., 2017;Tagliazucchi & Laufs, 2014). Therefore, the occasional falling asleep of adult participants in the scanner has been a problem for resting-state studies. To solve this issue, recent studies have shown that the use of non-social movies or videos helps to keep participants awake, increases compliance, and helps prevent social or emotional thoughts during mind-wandering (Anderson, Ferguson, Lopez-Larson, & Yurgelun-Todd, 2011;Cantlon & Li, 2013;Conroy, Singer, Guntupalli, Ramadge, & Haxby, 2013;Sabuncu et al., 2010). Likewise, previous studies have used non-social videos to acquire resting-state with fMRI in awake children (Müller, Kühn-Popp, Meinhardt, Sodian, & Paulus, 2015;Vanderwal, Kelly, Eilbott, Mayes, & Castellanos, 2015;Xiao, Friederici, Margulies, & Brauer, 2016). Furthermore, in adults, consistency within participants has been found between resting-state data acquired in a stimulus-free context and data acquired during observation of non-social videos, suggesting that observing such videos does not influence estimates of resting state connectivity significantly (Finn et al., 2017;Vanderwal et al., 2015).
As our knowledge of the development of the DMN thus far relies on data acquired from sleeping infants, it may possibly be incomplete. To compare infant and adult findings properly, restingstate data needs to be collected in awake infants. The current study aimed to fill this gap by investigating the developmental trajectory of connectivity within the DMN in awake infants. For this purpose, functional near-infrared spectroscopy (fNIRS) is a suitable neuroimaging method, as it is a non-invasive technique that measures changes in concentration in oxy-haemoglobin (HbO 2 ) and deoxyhaemoglobin (HHb) to index brain activation that can be used with awake infants (Elwell, 1995;Ferrari & Quaresima, 2012;Hoshi, 2016;Lloyd-Fox, Blasi, & Elwell, 2010;Wilcox & Biondi, 2015).
These characteristics, together with the fact that fNIRS is more robust to movement than other neuroimaging techniques, make this method highly suitable for acquiring resting-state recordings in infants under conditions similar to those typically used in studies with adults.
To our knowledge, only a few infant studies have measured spontaneous fluctuations in blood oxygenation during resting-state using fNIRS, but on sleeping participants (Homae et al., 2010;Konishi, Taga, Yamada, & Hirasawa, 2002;Taga et al., 2000). In particular, Homae et al. (2010) recorded resting-state in a longitudinal sample of sleeping infants at 4 days, at 3 and 6 months. An increase in functional connectivity was shown between the frontal, temporal, parietal and occipital regions. Additionally, while in the neonates connections were detected mainly within the same hemisphere, a more bilateral organisation of spontaneous networks emerged around the third month of life, when clusters of connections started to form across the midline (Homae et al., 2010). Recent adult studies have also used fNIRS to assess resting-state functional connectivity, suggesting it is a promising tool for this purpose (Lu et al., 2010;Mesquita, Franceschini, & Boas, 2010;Sasai et al., 2012). However, due to the fact that the near-infrared light can only penetrate a couple of centimetres into the scalp, its use is limited to the outer layers of the cortex. Therefore, in this study we measured connectivity between frontal, temporal, and parietal brain areas, which we will refer to as fronto-temporoparietal connectivity, as a proxy for the DMN. The approach of studying portions of the DMN as a proxy for this network has been recently adopted by adult studies, focusing in particular on the mPFC (Durantin, Dehais, & Delorme, 2015;Liang, Chen, Shewokis, & Getchell, 2016;Sasai et al., 2012) and the parietal lobes (Rosenbaum et al., 2017;Sasai et al., 2012).
To assess the developmental trajectory of fronto-temporoparietal connectivity, resting-state data were acquired with fNIRS in a longitudinal study at five time points. Participants were tested with the same resting-state procedure every 6 months, from 11 to 36 months. Regular intervals of data acquisition throughout the first 3 years of life allowed to capture the rapid neural development that takes place during this time (Johnson, 2001;Yamada et al., 1997). We hypothesised that we could find a gradual increase of fronto-temporoparietal connectivity over the first 3 years of life.
2 | METHODS 2.1 | Participants fNIRS resting-state data were acquired longitudinally when infants were 11, 18, 24, 30 and 36 months old. 1 Refer Table 1 for demographic information of the participants at each visit. All included infants were born full-term, healthy and with normal birth weight.
Written informed consent was obtained from the infant's caregiver prior to the start of the experiment.
Infants were excluded from the analysis if (a) their dataset did not reach the minimum length of 100 s of recording after behavioural coding (see section 2.5 for more details); (b) they refused to wear the NIRS hat or poor positioning of the NIRS hat; (c) more than 30% of the channels had to be excluded due to poor light intensity readings. Refer Table 1 (Everdell et al., 2005). Sampling rate of data acquisition was 10 Hz, and the mean power emitted by each laser diode was approximately 2 mW (Everdell et al., 2005).
At the 11th month visit, infants wore a custom-built headgear with a total of 30 channels. Data acquired at the other visits were collected using Easy Cap, caps made of soft black fabric, which provided a better fit on the participant's head, considering the increasing amount of hair (Figure 1a). At every visit, the custom-made NIRS array covered the temporal, parietal, and frontal areas bilaterally and two very similar designs were used to acquire data. The first array design

| Resting-state data acquisition
The resting-state acquisition took place in a dimly lit and sound attenuated room, with the infant sitting on their parent's lap at approximately 90 cm from a 117 cm plasma screen. The resting-state acquisition lasted until the participant became fussy, or until 6 min of data was recorded. To keep the infants awake and as still as possible, we showed them a screensaver-like video with colourful bubbles accompanied by relaxing music (Figure 2). The parent was asked not to talk during the experiment to avoid brain activation in areas of interest. If the parent talked to redirect the infant's attention to the screen or in case of fussiness or distraction, we excluded this chunk of data from the recording (see section 2.5 for more details).

| Co-registration of the fNIRS array
After the acquisition of the resting-state data, we logged the location of fNIRS array using the Polhemus Digitising System (http:// polhemus.com/scanning-digitizing/digitizing-products/) if the participant was still compliant, to allow us to co-register the fNIRS array on MRI structural scans. We registered five reference points (nasion, inion, right ear, left ear, Cz 3 ) and the location of the fNIRS optodes. In order to log the reference points and the optodes location as precise as possible, it was fundamental to keep the infants quiet and to limit their movements during the recording. Therefore, during the Polhemus recording, we showed them infant-friendly videos (e.g. clips from "In the Night Garden"). A marker placed on the back of the participant's cap allowed us to correct for head motion during the recording.
At every visit, we selected the best digitised recordings, based on the accuracy of the points marked in space compared to the optode locations in the pictures of the participant wearing the fNIRS cap that were taken after the recording (one from the front and two from the sides  (Fang & Boas, 2009), which estimates the paths of the photons from the source to the detector through the cortex. A cut-off of 25% of the voxels surrounding the spatial projection point was used to determine the anatomical label for each channel. Table 3 lists the anatomical labels (LPBA40 atlas) associated with each channel at each age belonging to the array design described in section 2.2. Table 4 describes channels belonging to each ROI at each age. Figure 3 provides a graphical representation of the brain areas covered by the fNIRS array used at every age, where the ROIs are colours coded. Figure 4 is a sensitivity map, showing brain regions that are sensitive to light attenuation changes given the fNIRS array we used in this work.
In this study, the connections between the frontal region and the temporoparietal region were defined as the connections between  (Yeo et al., 2011), but in relation to a more rostral portion of this region, which we did not cover with the array used in this study.

| Resting-state data pre-processing and analysis
Data analysis were carried out in MATLAB (MathWorks, Natick, MA). fNIRS resting-state data were extracted for each participant from all the channels and channels with mean intensity lower than 10 −3 μmol were excluded as such low intensity values indicate bad optode-scalp coupling ( Figure 5a). The global mean removal is a step which might be implemented in resting-state adult analysis. One of the most common method to perform it is the implementation of short-separation channels on the fNIRS cap (Brigadoi & Cooper, 2015;Gagnon et al., 2012). However, a recent infant study showed no differences between channels activations detected with and without the short-separation channels on 6-month-olds (Emberson, Crosswhite, Goodwin, Berger, & Aslin, 2016). Consistent with this, we have decided not to include this step in our pre-processing of the data.
Videos of the testing session were coded offline and periods where the infant moved, cried, or looked at something socially engaging (e.g., the mum or the experimenter) were marked as invalid, as well as periods during which the mum or experimenter were talking. To assess inter-coder reliability, 20% of the videos at every visit were blindly double-coded by another researcher. We found high reliability between the two coders (11 months, k = 0.78; 18 months, k = 0.84, 24 months, k = 0.85; 30 months, k = 0.89, 36 months, k = 0.80).
As Each functional connection between channels belonging to the frontal region and the temporoparietal region was inserted as a dependent variable in a linear mixed model (Verbeke & Molenberghs, 2000 This same procedure was used in other longitudinal studies that explored brain connectivity changes over time (Wierenga et al., 2018).

| Linear mixed model
To estimate how fronto-temporoparietal functional connectivity changes over the five visits, we analysed the fronto-temporoparietal connections in the HbO 2 signal using a linear mixed model.   Table 4). Among the 10 functional connections that showed a statistically significant change over time, only two did not survive the FDR correction for multiple comparisons (channel 27-channel 18 and channel 29-channel 10). As can be seen in Table 5 and in Figure 8a Regarding the linear mixed model performed on the ROIs, there was a statistically significant change with time in the mPFC-right TPJ functional connection-both when considering only channels 22 and 25 and also when considering the additional channels added from 24 months-with a maximum peak respectively at 18 and 24 months (Table 5 and Figure 8b). Both of these survived the FDR correction for multiple comparisons. There was also a marginal significant change in time in the mPFC-left TPJ connection (only channels 9 and 12), with a maximum peak at 24 months.

| Changes in connectivity outside the DMN over time
In order to assess whether the longitudinal variations that we observed in regions belonging to the DMN characterised this network only or were related to the rest of cortex as well, we assessed longitudinal changes intrahemispherically between IFG, MTG/STG and F I G U R E 5 (a) Representative segment of the resting-state raw data acquired. In the lower part of the figure, a red box marks channels that were excluded from the analysis because the mean intensity was lower than 10 −3 . On the remaining channels, red windows mark chunks of excluded data based on the behavioural coding. The grey windows represent the 8 s of additional data that was excluded after each invalid section.  Figure 9 and  (Horovitz et al., 2009;Mitra et al., 2017;Tagliazucchi & Laufs, 2014). Therefore, the use of fNIRS in awake infants has the potential of contributing to this investigation.
In the current study, we used fNIRS to explore the developmental  trajectory of fronto-temporoparietal connectivity-as a proxy of the DMN-in awake infants at five time points. We were able to acquire resting-state data in awake infants and toddlers, under conditions that are more similar to those typically used in adult studies, and we validated an analysis pipeline that can be applied in future studies. At every visit, we coregistered the fNIRS optodes on an MRI template of the same age. This allows us to more precisely estimate channels-brain correspondence and to adjust the channels-ROIs correspondence at every age, accounting for brain growth.
In line with previous resting state fNIRS studies (Lu et al., 2010;Sasai et al., 2011;White et al., 2009), the one sample t-tests at each visit on the HbO 2 and the HHb signals showed some consistency of the connectivity patterns that were measured in the two chromophores, suggesting that the data was reliable. Additionally, at every F I G U R E 8 (a) Functional connections that showed a significant change over time within the fronto-temporoparietal channels. (b) Graphical representation of the changes over time of the connections within the fronto-temporoparietal ROIs. **p < .05 that survived the FDR correction for multiple comparisons, *p < .05; † p < .065 age connectivity reached stability in most of the infants after about 60-90 s of data included, which is consistent with a recent study on children showed that as little as 1 min of resting-state fNIRS recording is sufficient to obtain accurate functional connectivity estimation (Wang, Dong, & Niu, 2017).

| Changes in the fronto-temporoparietal functional connectivity over time
Results from the linear mixed model analysis between the frontotemporoparietal channels and between fronto-temporoparietal ROIs showed stronger fronto-temporoparietal connections at older ages compared to younger ages, consistent with previous studies that have found a gradual increase of DMN connectivity over the first years of life Gao et al., 2009). Results showed a maximum increase of the functional connections at 24 and 30 months compared to the 11th month visit. One may think that the maximum peak of the functional connectivity change at 24 and 30 months could be related to methodological aspects, such as a higher level of noise in the data at these ages, as a high level of movement during the resting-state acquisition would have most likely led to a spurious increase in functional connectivity (Deen & Pelphrey, 2012;Power, Barnes, Snyder, Schlaggar, & Petersen, 2012;van Dijk, Sabuncu, & Buckner, 2012). However, we took great care to remove sections of the data affected by motion artefacts during the pre-F I G U R E 9 Functional connections outside the DMN that showed a significant change over time (left and right hemisphere) and interhemispheric connections between homologous regions. **p < .05 that survived the FDR correction for multiple comparisons, *p < .05; † p < .065 T A B L E 6 Changes of the connections outside the DMN and interhemispheric connectivity over time processing. When the infants were 11 months, they provided the noisiest dataset, and the quality of the resting-state recordings increased with age. More likely, the maximum increase of the functional connections at 24 and 30 months compared to the 11 month visit indicates a stability in the strength of the connectivity in DMN regions, which is consistent with a previous study by Gao et al. (2009), that showed that by 2 years, the DMN is functionally similar to the DMN observed in adults, with long-range connections between the frontal cortex and the posterior regions of the DMN (Gao et al., 2009(Gao et al., , 2016. In this study, we reported major changes in DMN connectivity between the first and the second year of life, which is consistent with the remarkable increments in white matter tracts connecting core hubs of the DMN that have been documented within this period (Fan et al., 2011). However, Gao et al. (2009Gao et al. ( , 2015, showed limited variations in the DMN functional connectivity within this age when using fMRI with sleeping infants. The DMN connectivity estimated at 24 months similar to the network observed in adults is mainly driven by changes up to the first year of life, followed by minor increases between the first and the second year (Gao et al., 2009(Gao et al., , 2015. One may wonder whether these dissimilarities are due to differences in arousal states of the participants. It is difficult to draw clear comparisons here, as we have not collected resting-state data from infants younger than 1 year of age. Future research could assess functional connectivity between DMN regions in very young awake infants using fNIRS, as we have done in this study. This would allow us to better understand whether the major variations in DMN connectivity reported in different periods in the current work and previous studies (Gao et al., 2009(Gao et al., , 2015 can be explained by the effect of sleep on functional connectivity or there are additional reasons. The peak in fronto-temporoparietal connectivity at 24-30 months seems to be followed by a decrease. This nonlinear development could be explained by pruning processes, that is, the removal of redundant connections (Huttenlocher, Vasilyeva, & Shimpi, 2004), resulting in a more efficient set of connections (Thompson et al., 2005) and enabling the reorganisation of functional networks (Gao et al., 2016;Levitt, 2003). Pruning is known to be a region-specific process, affecting different brain regions at different stages of the development (Casey, Tottenham, Liston, & Durston, 2005). The increase in connections and the subsequent pruning happens last in the frontal lobe, while this process affects other regions such as the auditory, the visual and the sensorimotor cortex at an earlier age (Huttenlocher & Dabholkar, 1997). Consistent with our findings, it has been shown that a peak in synaptic density in the frontal cortex is achieved only after the first year of life (Huttenlocher & Dabholkar, 1997;Tierney & Nelson, 2009) with pruning of frontal connections starting at around 2 years (Casey et al., 2005;Kolb & Gibb, 2011). Patterns of decreases following increases in the maturity of the brain have been observed not only in relation to functional connectivity.
Other works documented the same non-linear growth in cortical thickness (Shaw et al., 2008), in some white matter tracts (Mukherjee et al., 2001), and in grey matter density (Sowell, Thompson, & Toga, 2004;Toga, Thompson, & Sowell, 2006), findings which are consistent with a reorganisation process in the brain after its growth for a more efficient activity. However, why we observed this non-linear growth particularly in the fronto-temporoparietal network rather than in the connections outside the DMN is unclear. In this respect, it is important to point out that most of the previous longitudinal studies acquired resting-state data up to 2 years, or with intervals not as frequent as 6 months Gao et al., 2009;Homae et al., 2010), or from 6 to 7 years of age to adulthood (Jolles, Van Buchem, Crone, & Rombouts, 2011;Marusak et al., 2017;Supekar et al., 2010;Supekar, Musen, & Menon, 2009)  A gradual increase in fronto-temporoparietal connectivity until to 24 months of age, particularly significant in the right hemisphere, is consistent with the hypothesised relationship between the DMN and self-processing (Buckner & Carroll, 2007;Golland et al., 2008;Qin & Northoff, 2011). The study of the development of the sense of self has recently been a topic of much interest in developmental psychology and whether we currently have appropriate measurements for its assessment is under debate. The sense of self is thought to develop between 18 and 24 months of age (Amsterdam, 1972;Rochat, 2003), and it is typically assessed using the mirror self-recognition task (Amsterdam, 1972). However, there is no general consensus on the significance of mirror self-recognition (for some criticisms, see Heyes & Swartz, 1997;Mitchell, 1993 (Homae, Watanabe, & Taga, 2016).
Interhemispheric connectivity between homologous regions seems to significantly change after the first year of life, and then reaches stability. These results seem to suggest that the pattern of longitudinal changes observed in the DMN is specific to this network.
However, as the fNIRS method only allows us to measure the cortical surface, and our optodes did not cover the entire head, we acknowledge that these additional analyses do not allow us to completely rule out potential changes in general brain maturation accounting for some of the developmental change we observed in the DMN. The only way to correctly assess change in connectivity in the entire brain would be to acquire MRI images. One might think that the presence of interhemispheric connectivity as early as 11 months is in contrast with the literature about the protracted development of the corpus collosum (Pujol, Vendrell, Junqué, Martí-Vilalta, & Capdevila, 1993). However, some white matter tracts in the corpus callosum are present right after birth, and interhemispheric connectivity before the first year of life has been documented elsewhere (Gao et al., 2009;Homae et al., 2010;Keehn, Wagner, Tager-Flusberg, & Nelson, 2013;Perani et al., 2011;Smyser, Snyder, & Neil, 2011;Taga et al., 2011). The corpus collosum undergoes a slow continuous development from infancy until early adulthood (for example see Chavarria, Sánchez, Chou, Thompson, & Luders, 2014;Giedd et al., 1996Giedd et al., , 1999Giorgio et al., 2010;Hinkley et al., 2012;Keshavan et al., 2002;Luders, Thompson, & Toga, 2010), but this does not mean that interhemispheric connectivity is not present before the corpus callosum is completely mature.

| Methodological limitations and further considerations
It may be interesting to notice that the main reason for exclusion from the analysis changed over time. While participants at younger ages were mainly excluded because their artefact-free resting-state data did not reach the minimum required length or because they refused to wear the fNIRS cap, at older ages the main reason for exclusion was the high number of channels with poor light intensity. The difficulty for young infants to reach a quiet state (especially after having already been presented with several other experiments) significantly reduced amount of data available for analysis at the 11th month visit compared with the ones, limiting the longitudinal comparisons.
In addition, testing participants at different visits with different fNIRS array configurations restricted the comparisons of some of the fronto-temporoparietal connections between all the visits. Although the 44-channel configuration was an extension of the 30-channel configuration, comparisons at 11 and 18 months were limited by the absence of the additional channels. The reason for adding the additional channels to the 30-channel configuration was to improve the detection of TPJ spontaneous fluctuations, one of the core regions for understanding developmental changes in the fronto-temporoparietal connections. However, it is important to highlight that changes over time between the mPFC-right TPJ connectivity showed the same significant pattern when considering the TPJ channels belonging to 30-channel configuration only, and when we consider those added in the 44-channel configuration, suggesting that this result is not driven by the extension of the spatial coverage the right TPJ region.
At every visit, we have co-registered the NIRS array of a subset of participants with MRI scans, but we do acknowledge that we have not accounted for inter-individual differences of each participant at each age point. However, we have excluded every infant with poor position of the NIRS headband/cap, based on the pictures taken during the testing sessions. Without individual MRI scans, it is difficult to estimate for each participant whether we measure from exactly the same area, which is one of the main limitations of fNIRS. We believe that by taking a representative sample of infants and doing the coregistration based on MRI templates that closely matched their head shape and size, the ROIs at every visit were likely to be accurate for the majority of participants.
In the current study, we measured the development of frontotemporoparietal connectivity while the participants watched a screen-saver like video. Although the use of non-social videos to measure resting state connectivity has been previously validated with children and adults (Müller et al., 2015;Vanderwal et al., 2015;Xiao et al., 2016), and consistency in functional connectivity estimated during non-social videos and rest has been documented (Finn et al., 2017;Vanderwal et al., 2015), we are aware that the presence of audiovisual stimuli does not entirely equate the testing conditions of resting-state studies historically performed with adults. However, the use of this screensaver video was the only feasible way to measure resting state connectivity with infants and toddlers while they were awake. In fact, given that sleep stages (Mitra et al., 2017;Tagliazucchi & Laufs, 2014) and movement (Power et al., 2012;van Dijk et al., 2012) affect connectivity estimation, our main priority was that the participants remained calm and awake during the recording, and the videos helped with this. On the other side, we would like to highlight how the absence of background noise and the limited physical constraints which characterise a fNIRS lab setting rather than a fMRI one could be ideal for resting-state studies, where it is important to interfere as little as possible with participants' mind-wandering. In fact, because there is no requirement for participants to lay or sit perfectly still as in fMRI studies, fNIRS allows for the recording of resting state data in a more naturalistic setting. In this respect, researchers are currently working towards making this technology wireless and improving the flexibility and the comfort of the caps even further (Pinti et al., 2015;Zhao et al., 2019), aspects that future resting-state studies can benefit from.

| CONCLUSIONS
This is the first time that functional connectivity was estimated longitudinally in awake infants with fNIRS. Our results suggest a gradual increase of fronto-temporoparietal connectivity over the first years of life, with a peak at 24 and 30 months which might indicate that by this age the DMN is fully developed. There seems to be a slight decrease after this point, which might be consistent with the process of connections pruning, starting at 2 years of age. From a methodological point of view, this study proposes a novel method of resting-state data acquisition with awake infants, and provides a data analysis pipeline for the investigation of functional connectivity, which will facilitate the advancement of research in this field. We hope that fNIRS researchers interested in exploring functional connectivity in awake infants can benefit from this work.

ACKNOWLEDGMENTS
We are very grateful to all the infants and parents who participated in this study.